Re: [pcp] pcp update: fetchgroups v4: with event-field support

[Nathan Scott <nathans@xxxxxxxxxx>]

Oh, and...

----- Original Message -----
> 
> > Remember, the point of sentinel values is to be convenient for
> > computation & output, should the application *not care* [...]
> 
> Firstly, reporting incorrect information is a big problem.  And zero
> as a numeric value is so often of meaning & importance that no matter
> how uncaring a tool author is about ENODATA, we cannot encourage it.
> [...]
> Secondly, above rationale is [] inconsistent with the approach taken
> for double/float which generates NaN instead of 0.0.  We need to pick
> one or the other, not half-and-half, and its important to not provide
> potentially-very-misleading zero values as defaults.
> 

Thirdly, zeroed metric values are a wonderful source of divide-by-zero
errors causing lazily-written applications to fail - these will end up
needing special-case handling that other sentinel values wouldn't (and
-1 has nice propagating properties as an alternative there too).

cheers.

--
Nathan